red personal

Esta página contiene el código para generar análisis de redes personales (ego networks) en Twitter.

Set up

library(rtweet)
source("createTokens.R")  ## keys y tokens privados
source("rtweet_functions.R") ## funciones para trabajar con múltiples tokens

library(tidyverse)
library(igraph)
library(tidygraph)
library(ggraph)
library(ggwordcloud)
library(tidytext)
theme_set(theme_custom())

El primer paso consiste en escoger un usuario focal (o “ego”) a partir del cual construímos una red personal.

ego <- "sergemont" # 
ego_info <- lookup_users(ego, token = sample(token, 1))

ego_info$followers_count

## [1] 8673

Nombre: Sergio Montero

Usuario: sergemont

Seguidores: 8673

Amigos: 2419

Se unió a Twitter en 2012-09-09 14:19:18

Este análisis está dividido en tres partes.

1. La red de seguidores del usuario focal
2. La red de “amigos” del ususario focal
3. La red de amigos-seguidores del usuario focal

Cada una de estas tres dimensiones corresponde a flujos de interacción diferentes. La primera consiste de los usuarios que reciben información de sergemont, la segunda son los usuarios que generan la información recibida por sergemont, y la tercera consiste en los usuarios donde el flujo de información es recíproco.

Este código es de acceso libre excepto por los keys y tokens privados que se consiguen abriendo una cuenta de desarrollador en https://developer.twitter.com/

Red de seguidores

El siguiente código extrae la lista de seguidores de sergemont (cada uno identificado con un user_id).

ego_followers <- get_followers(ego, token = sample(token, 1), n = ego_info$followers_count)
ego_followers

## # A tibble: 8,673 x 1
##    user_id            
##    <chr>              
##  1 120180081          
##  2 68043845           
##  3 1330931515224715265
##  4 94231765           
##  5 122851379          
##  6 803399965          
##  7 1019674215769726976
##  8 192409907          
##  9 256959302          
## 10 1326849155243839490
## # … with 8,663 more rows

Este user_id es exclusivo a cada cuenta, incluso cuando el usuario decide cambiar su nombre.

El siguiente código crea una carpeta llamada *_friends_of_followers/ donde queda archivado la lista de los seguidores de cada uno de estos usuarios.

Dependiendo del número de usuarios y el número de Tokens, esto puede llegar a demorarse varias horas (o incluso días).

outfolder <- paste0(ego, "_friends_of_followers/")
if (!dir.exists(outfolder)) dir.create(outfolder)

users_done <- str_replace(dir(outfolder), ".rds", "")
users_left <- setdiff(ego_followers$user_id, users_done)

while (length(users_left) > 0) { 
  
  new_user <- users_left[[1]]
  
  friends_of_user <- try(multi_get_friends(new_user, token))
  
  file_name <- str_glue("{outfolder}{new_user}.rds")
  write_rds(friends_of_user, file_name, compress = "gz")
  users_left <- users_left[-which(users_left %in% new_user)] ## int. subset
  
}

Para algunos usuarios esta información es imposible de conseguir porque son cuentas protegidas.

En este caso, no se puede obtener información sobre el 7.5% de los sequidores de sergemont.

Edge list

Para construir la red, tomamos toda la lista de usuarios y sus amigos y los organizamos en dos columnas, donde cada fila indica un usario (from) siguiendo a otro usario (to).

edge_list <- list.files(outfolder, full.names = TRUE) %>% 
  map(read_rds)
  
edge_list <- edge_list[-error_index] %>% 
  bind_rows() 

edge_list

## # A tibble: 12,073,379 x 2
##    from                to                 
##    <chr>               <chr>              
##  1 1000056137255825410 1325845228398342144
##  2 1000056137255825410 28221296           
##  3 1000056137255825410 707384515979706368 
##  4 1000056137255825410 87697296           
##  5 1000056137255825410 252589325          
##  6 1000056137255825410 903239720          
##  7 1000056137255825410 189385477          
##  8 1000056137255825410 563225180          
##  9 1000056137255825410 710121377391628292 
## 10 1000056137255825410 1472812134         
## # … with 12,073,369 more rows

Aquí hay 12,073,379 conexiones. Sin embargo, aquí están incluídos conexiones on usuarios más allá de los que siguen a sergemont.

ego_followers_info <- lookup_users(ego_followers$user_id, token = sample(token), 1)
write_rds(ego_followers_info, paste0(ego, "_follower_info.rds"), compress = "gz")

También podemos conseguir metadatos sobre cada usuario.

ego_followers_info <- read_rds(paste0(ego, "_follower_info.rds")) %>% 
  filter(!protected) %>% 
  select(
    user_id, screen_name, lang, name, location, description,
    ends_with("count"), -starts_with("quote"), 
    -starts_with("retweet"), -reply_count,
    -starts_with("fav")
    ) %>% 
    rename(name = user_id, user_name = name)

id_dict <- ego_followers_info %>% 
  select(name, screen_name) %>% 
  deframe()

Por ejemplo, esta es la información que corresponde a los seguidores de sergemont con mayor número de seguidores.

ego_followers_info %>% 
  arrange(desc(followers_count)) %>% 
  select(screen_name, description, location, followers_count, friends_count)

## # A tibble: 8,009 x 5
##    screen_name  description            location    followers_count friends_count
##    <chr>        <chr>                  <chr>                 <int>         <int>
##  1 miguelhotero "Presidente Editor Di… "El Nacion…         1804453       1064718
##  2 BogotaET     "Información del acon… "Bogotá"            1121542          1444
##  3 Rmayorga     "Periodista univallun… "ÜT: 4.663…          597097          2706
##  4 AngelicaLoz… "Ciudadana, senadora … ""                   464197         16375
##  5 ClaMoralesM  "Periodista. \nColumn… "Armenia"            414681          2517
##  6 VanessaRosa… "Escritora • Crítica … "Bogotá, C…          299383          1565
##  7 mapatilla    "Escribo punto al fin… "Colombia"           253829          1417
##  8 VivaAirCol   "Somos los más puntua… "Colombia"           230489         38883
##  9 JuanSheput   "Político e ingeniero… "Lima, Per…          196957          2614
## 10 Uniandes     "Universidad de los A… "Bogotá, C…          140972           162
## # … with 7,999 more rows

Finalmente nos interesa la red personal de seguidores de sergemont, por lo cual eliminamos las conexiones entre usuarios que se encuentran por fuera de sus 8673

edge_list <- edge_list %>% 
  filter(to %in% ego_followers_info$name) %>% 
  filter(from %in% ego_followers_info$name)

edge_list

## # A tibble: 437,424 x 2
##    from                to                 
##    <chr>               <chr>              
##  1 1000056137255825410 15432179           
##  2 1000056137255825410 848923514119884800 
##  3 1000056137255825410 2473739315         
##  4 1000056137255825410 142240353          
##  5 1000056137255825410 397366316          
##  6 1000224690491871232 82742830           
##  7 1000224690491871232 1104834306         
##  8 1000224690491871232 1148049055341326336
##  9 1000224690491871232 167394234          
## 10 1000224690491871232 37193785           
## # … with 437,414 more rows

La red personal de seguidores de sergemont que pudimos reconstruir tiene 8009 usuarios con 437424 conexiones.

Red Personal

ego_network <- edge_list %>% 
  tidygraph::as_tbl_graph() %>% 
  left_join(ego_followers_info) %>% 
  rename(name = screen_name, user_id = name) %>% 
  select(name, everything())

ego_network

## # A tbl_graph: 7857 nodes and 437424 edges
## #
## # A directed simple graph with 1 component
## #
## # Node Data: 7,857 x 10 (active)
##   name  user_id lang  user_name location description followers_count
##   <chr> <chr>   <chr> <chr>     <chr>    <chr>                 <int>
## 1 Mago… 100005… es    Magove    ""       ""                      101
## 2 Juan… 100022… es    JuanitoM… "Bogotá… ""                       69
## 3 Vale… 100022… es    Valeria   "Colomb… "Derecho. …             134
## 4 juli… 100037… es    J U L I … "Bogotá… ""                     2536
## 5 Gafa… 100044… es    Jesús Da… "Monter… "Política-…              57
## 6 susa… 100047… pt    Susana Y… ""       ""                       33
## # … with 7,851 more rows, and 3 more variables: friends_count <int>,
## #   listed_count <int>, statuses_count <int>
## #
## # Edge Data: 437,424 x 2
##    from    to
##   <int> <int>
## 1     1  2304
## 2     1  7122
## 3     1  3592
## # … with 437,421 more rows

## Estadísticas descriptivas

ego_network <- ego_network %>% 
  mutate(
    out_degree = centrality_degree(mode = "out"),
    in_degree = centrality_degree(mode = "in"),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  )

ego_network 

## # A tbl_graph: 7857 nodes and 437424 edges
## #
## # A directed simple graph with 1 component
## #
## # Node Data: 7,857 x 15 (active)
##   name  user_id lang  user_name location description followers_count
##   <chr> <chr>   <chr> <chr>     <chr>    <chr>                 <int>
## 1 Mago… 100005… es    Magove    ""       ""                      101
## 2 Juan… 100022… es    JuanitoM… "Bogotá… ""                       69
## 3 Vale… 100022… es    Valeria   "Colomb… "Derecho. …             134
## 4 juli… 100037… es    J U L I … "Bogotá… ""                     2536
## 5 Gafa… 100044… es    Jesús Da… "Monter… "Política-…              57
## 6 susa… 100047… pt    Susana Y… ""       ""                       33
## # … with 7,851 more rows, and 8 more variables: friends_count <int>,
## #   listed_count <int>, statuses_count <int>, out_degree <dbl>,
## #   in_degree <dbl>, betweenness <dbl>, authority_score <dbl>,
## #   eigen_centrality <dbl>
## #
## # Edge Data: 437,424 x 2
##    from    to
##   <int> <int>
## 1     1  2304
## 2     1  7122
## 3     1  3592
## # … with 437,421 more rows

La siguiente gráfica muestra la influencia de cada usuario en Twitter (eje horizontal) vs la influencia de cada usuario dentro de la red personal de seguidores (eje vertical)

ego_network %>% 
  as_tibble() %>% 
  #filter(in_degree > 5) %>% 
  ggplot(aes(followers_count, in_degree)) + 
  geom_point() 

ego_network %>% 
  as_tibble() %>% 
  mutate(label_name = ifelse(
    test = rank(-followers_count) <= 10 | rank(-in_degree) <= 10, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(followers_count, in_degree)) + 
  geom_point() + 
  ggrepel::geom_label_repel(aes(label = label_name), size = 3)

Clusters

set.seed(123)
clusters <- igraph::cluster_walktrap(graph = ego_network, steps = 7)

cluster_df <- tibble(cluster = factor(clusters$membership), name = clusters$names) 
  
cluster_df <- cluster_df %>% 
  group_by(cluster) %>% 
  filter(n() >= 10) %>% 
  ungroup()

ego_network <- ego_network %>% 
  left_join(cluster_df)

ego_network %>% 
  as_tibble() %>% 
  arrange(desc(in_degree)) %>% 
  filter(!is.na(cluster)) %>% 
  group_by(cluster) %>%
  filter(rank(-authority_score) <= 30) %>% 
  ggplot(aes(label = name, size = log(in_degree), color = in_degree)) + 
  geom_text_wordcloud_area(family = "Avenir Next Condensed") + 
  facet_wrap(~cluster) + 
  labs(title = "Seguidores prominentes en cada cluster") + 
  scale_color_gradient(low = "grey", high = "purple") 

Tamaño de cada cluster:

ego_network %>% as_tibble() %>% count(cluster)

## # A tibble: 9 x 2
##   cluster     n
##   <fct>   <int>
## 1 1         380
## 2 2         162
## 3 3        2181
## 4 4         249
## 5 5         311
## 6 6        3941
## 7 7         338
## 8 11        129
## 9 <NA>      166

¿Quiénes son los usuarios que funcionan como “puentes”?

ego_network %>% 
  as_tibble() %>% 
  arrange(desc(betweenness)) %>% 
  select(name, description, location)

## # A tibble: 7,857 x 3
##    name        description                                    location          
##    <chr>       <chr>                                          <chr>             
##  1 malbarracin "Santandereano en el exilio. Abogado, activis… "Bogotá"          
##  2 AngelicaLo… "Ciudadana, senadora de Colombia 🇨🇴 Partido V… ""                
##  3 psanabria   "Public Policy & Management Scholar | Profeso… "Latin America"   
##  4 JuanitaGoe  "Representante a la Cámara por Bogotá (2018-2… "Bogotá, D.C., Co…
##  5 ErikVergel  "PhD City Regional Planning @UNC Master Urban… "Bogotá, D.C., Co…
##  6 LaCiudadVe… "Promovemos ciudades y regiones más sostenibl… "América Latina"  
##  7 riveraalza… "Concejal de Bogotá. Alianza Verde. Gestor Ur… "Bogotá"          
##  8 Afbrico     "#ElProfeDeFút⚽\n\nLas opiniones acá expresad… "Polombia"        
##  9 miguelhote… "Presidente Editor Diario El Nacional"         "El Nacional, Los…
## 10 Rivas_Sant… "Director/presentador de Puntos Capitales. Pa… ""                
## # … with 7,847 more rows

cols <- c("betweenness", "in_degree", "out_degree", "followers_count", "friends_count")

ego_network %>% 
  as_tibble() %>% 
  group_by(cluster) %>% 
  summarize(across(all_of(cols), mean)) %>% 
  arrange(desc(betweenness))

## # A tibble: 9 x 6
##   cluster betweenness in_degree out_degree followers_count friends_count
##   <fct>         <dbl>     <dbl>      <dbl>           <dbl>         <dbl>
## 1 1            20153.     54.1       51.8            1945.         1563.
## 2 3            15765.     67.4       68.7            1825.         1717.
## 3 2            13579.     42.4       40.3            1712.         1775.
## 4 4            12667.     35.2       37.9            1677.         1776.
## 5 6            12059.     54.4       54.4            2368.         1388.
## 6 5             7662.     31.4       31.2            1719.         1864.
## 7 7             6854.     65.9       60.6            1567.         1520.
## 8 11            4923.     57.6       56.1            1069.         1157.
## 9 <NA>          1640.      1.52       1.81           1603.         1427.

Subset

Dada la información anterior podemos enfocarnos en segmentos particulares de la red personal.

Por ejemplo, podemos enfocarnos exclusivamente en los usuarios que hacen parte de los grupos etiquetados con 12 y 4.

ego_network_subset <- ego_network %>% 
  filter(cluster %in% c(12, 4, 2, 6)) %>% 
    mutate(
    out_degree = centrality_degree(mode = "out"),
    in_degree = centrality_degree(mode = "in"),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  ) 

ego_network_subset %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = in_degree), 
                  shape = 21, color = "white", show.legend = FALSE) 

ego_network_subset %>% 
  as_tibble() %>% 
  mutate(label_id = ifelse(
    test = rank(-betweenness) <= 10 |rank(-in_degree) <= 10, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(betweenness, in_degree, color = cluster)) +
  geom_point() +
  ggrepel::geom_label_repel(aes(label = label_id), size = 3)

ego_network_subset %>% 
  group_by(cluster) %>% 
  mutate(label_name = ifelse(
    test = rank(-authority_score) <= 5 | rank(-betweenness) <= 5,
    yes = name,
    no = NA_character_
  )) %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = betweenness), 
                  shape = 21, color = "white", show.legend = FALSE) +
  geom_node_label(aes(label = label_name), 
                  repel = TRUE, alpha = 3/4, size = 3) 

Red de amigos

Esta sección repite el análisis anterior para la red personal de amigos de sergemont

outfolder <- paste0(ego, "_friends_of_friends/")
if (!dir.exists(outfolder)) dir.create(outfolder)

ego_friends <- get_friends(ego, token = sample(token, 1))
ego_friends

## # A tibble: 2,419 x 2
##    user      user_id            
##    <chr>     <chr>              
##  1 sergemont 66676632           
##  2 sergemont 599551829          
##  3 sergemont 1219662920700911616
##  4 sergemont 19128902           
##  5 sergemont 3149487047         
##  6 sergemont 734748619861983232 
##  7 sergemont 419533496          
##  8 sergemont 95654761           
##  9 sergemont 1315224240304648194
## 10 sergemont 1318461879409401859
## # … with 2,409 more rows

users_done <- str_replace(dir(outfolder), ".rds", "")
users_left <- setdiff(ego_friends$user_id, users_done)

while (length(users_left) > 0) { 
  
  new_user <- users_left[[1]]
  
  friends_of_user <- try(multi_get_friends(new_user, token))
  
  file_name <- str_glue("{outfolder}{new_user}.rds")
  write_rds(friends_of_user, file_name, compress = "gz")
  users_left <- users_left[-which(users_left %in% new_user)] ## int. subset
  
}

En este caso, no se puede obtener información sobre el 0.2% de los amigos de sergemont.

Edge list

edge_list <- list.files(outfolder, full.names = TRUE) %>% 
  map(read_rds)
  
edge_list <- edge_list[-error_index] %>% bind_rows()

edge_list

## # A tibble: 3,989,460 x 2
##    from                to                 
##    <chr>               <chr>              
##  1 1000369692278738944 19103793           
##  2 1000369692278738944 980078479269089280 
##  3 1000369692278738944 99378819           
##  4 1000369692278738944 13876622           
##  5 1000369692278738944 1123725326817873920
##  6 1000369692278738944 21577894           
##  7 1000369692278738944 326619024          
##  8 1000369692278738944 1002300185391710208
##  9 1000369692278738944 344103743          
## 10 1000369692278738944 2347372296         
## # … with 3,989,450 more rows

ego_friends_info <- lookup_users(ego_friends$user_id, token = token)
write_rds(ego_friends_info, paste0(ego, "_friends_info.rds"), compress = "gz")

ego_friends_info <- read_rds(paste0(ego, "_friends_info.rds")) %>% 
  filter(!protected) %>% 
  select(
    user_id, screen_name, lang, name, location, description,
    ends_with("count"), -starts_with("quote"), 
    -starts_with("retweet"), -reply_count,
    -starts_with("fav")
    ) %>% 
    rename(name = user_id, user_name = name)

id_dict <- ego_friends_info %>% 
  select(name, screen_name) %>% 
  deframe()

Esta es la información que corresponde a los amigos de sergemont con mayor número de seguidores.

ego_friends_info %>% 
  arrange(desc(followers_count)) %>% 
  select(screen_name, description, location, followers_count, friends_count)

## # A tibble: 2,398 x 5
##    screen_name  description             location   followers_count friends_count
##    <chr>        <chr>                   <chr>                <int>         <int>
##  1 TheEconomist "News and analysis wit… London            25235201           127
##  2 BernieSande… "U.S. Senator for Verm… Vermont           13885863          1479
##  3 theweeknd    "https://t.co/OP85ArZo… Toronto, …        13542464             1
##  4 KamalaHarris "Vice President-Elect … California        12297917           736
##  5 AOC          "US Representative,NY-… Bronx + Q…        10600561          2914
##  6 guardian     "The need for independ… London             9319045          1082
##  7 NewYorker    "Unparalleled reportin… New York,…         8973522           373
##  8 el_pais      "La mejor información … Madrid             7843092           779
##  9 AJEnglish    "Hear the human story … Doha, Qat…         6513223           221
## 10 YourAnonCen… "Exposing Human Rights… in your a…         5959311           755
## # … with 2,388 more rows

edge_list <- edge_list %>% 
  filter(to %in% ego_friends_info$name) %>% 
  filter(from %in% ego_friends_info$name)

edge_list

## # A tibble: 203,000 x 2
##    from                to        
##    <chr>               <chr>     
##  1 1000369692278738944 63044418  
##  2 1000369692278738944 14700117  
##  3 1000369692278738944 842242609 
##  4 1000369692278738944 1161084242
##  5 1000369692278738944 111619549 
##  6 1000369692278738944 80908670  
##  7 1000369692278738944 14362241  
##  8 1000369692278738944 24889586  
##  9 1000369692278738944 200514238 
## 10 1000369692278738944 3316287402
## # … with 202,990 more rows

La red personal de seguidores de sergemont que pudimos reconstruir tiene 2398 usuarios con 203000 conexiones.

Red Personal

ego_network <- edge_list %>% 
  tidygraph::as_tbl_graph() %>% 
  left_join(ego_friends_info) %>% 
  rename(name = screen_name, user_id = name) %>% 
  select(name, everything())

## Estadísticas descriptivas

ego_network <- ego_network %>% 
  mutate(
    out_degree = centrality_degree(mode = "out"),
    in_degree = centrality_degree(mode = "in"),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  )

ego_network

## # A tbl_graph: 2394 nodes and 203000 edges
## #
## # A directed simple graph with 1 component
## #
## # Node Data: 2,394 x 15 (active)
##   name  user_id lang  user_name location description followers_count
##   <chr> <chr>   <chr> <chr>     <chr>    <chr>                 <int>
## 1 TheW… 100036… en    The War … "Brookl… "A podcast…           19565
## 2 digi… 100125… en    Digital … "on the… "Building …            3442
## 3 Vall… 100138… es    Vallekas… "Valle … "Cuenta of…            2074
## 4 UoMP… 100140… en    Policy@M… ""       "Influence…           11634
## 5 Sani… 100163… es    Carolina… ""       "«Si a mí …          110630
## 6 RapP… 100257… es    RAP Pací… ""       "Entidad p…            1435
## # … with 2,388 more rows, and 8 more variables: friends_count <int>,
## #   listed_count <int>, statuses_count <int>, out_degree <dbl>,
## #   in_degree <dbl>, betweenness <dbl>, authority_score <dbl>,
## #   eigen_centrality <dbl>
## #
## # Edge Data: 203,000 x 2
##    from    to
##   <int> <int>
## 1     1  1852
## 2     1   470
## 3     1  2127
## # … with 202,997 more rows

La siguiente gráfica muestra la influencia de cada usuario en Twitter (eje horizontal) vs la influencia de cada usuario dentro de la red personal de amigos (eje vertical)

ego_network %>% 
  as_tibble() %>% 
  #filter(in_degree > 5) %>% 
  ggplot(aes(followers_count, in_degree)) + 
  geom_point() 

ego_network %>% 
  as_tibble() %>% 
  mutate(label_name = ifelse(
    test = rank(-followers_count) <= 10 | rank(-in_degree) <= 10, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(followers_count, in_degree)) + 
  geom_point() + 
  ggrepel::geom_label_repel(aes(label = label_name), size = 3)

Clusters

clusters <- igraph::cluster_walktrap(graph = ego_network, steps = 12)

cluster_df <- tibble(cluster = factor(clusters$membership), name = clusters$names) 
  
cluster_df <- cluster_df %>% 
  group_by(cluster) %>% 
  filter(n() >= 10) %>% 
  ungroup()

ego_network <- ego_network %>% 
  left_join(cluster_df)

ego_network %>% 
  as_tibble() %>% 
  arrange(desc(in_degree)) %>% 
  filter(!is.na(cluster)) %>% 
  group_by(cluster) %>%
  filter(rank(-authority_score) <= 30) %>% 
  ggplot(aes(label = name, size = log(in_degree), color = in_degree)) + 
  geom_text_wordcloud_area(family = "Avenir Next Condensed") + 
  facet_wrap(~cluster) + 
  labs(title = "Amigos prominentes en cada cluster") + 
  scale_color_gradient(low = "grey", high = "purple") 

Tamaño de cada cluster:

ego_network %>% as_tibble() %>% count(cluster)

## # A tibble: 4 x 2
##   cluster     n
##   <fct>   <int>
## 1 1         888
## 2 2          50
## 3 3         538
## 4 4         918

¿Quiénes son los usuarios que funcionan como “puentes”?

ego_network %>% 
  as_tibble() %>% 
  arrange(desc(betweenness)) %>% 
  select(name, description, location)

## # A tibble: 2,394 x 3
##    name        description                                      location        
##    <chr>       <chr>                                            <chr>           
##  1 ErikVergel  PhD City Regional Planning @UNC Master Urban Ma… "Bogotá, D.C., …
##  2 spaceforpe… political geographer in Peace Studies at Kent S… "Kent, OH"      
##  3 AngelicaLo… Ciudadana, senadora de Colombia 🇨🇴 Partido Verd… ""              
##  4 malbarracin Santandereano en el exilio. Abogado, activista … "Bogotá"        
##  5 antipodeon… Antipode: A Radical Journal of Geography, publi… ""              
##  6 psanabria   Public Policy & Management Scholar | Profesor e… "Latin America" 
##  7 regstud     Supporting and connecting researchers and polic… "Brighton, Engl…
##  8 UrbanTheor… This is the twitter platform of Neil Brenner, c… "Chicago"       
##  9 CiderUnian… Twitter oficial del Centro Interdisciplinario d… "Bogotá - Colom…
## 10 PPS_Placem… We bring community-powered public spaces to lif… "New York, NY"  
## # … with 2,384 more rows

cols <- c("betweenness", "in_degree", "out_degree", "followers_count", "friends_count")

ego_network %>% 
  as_tibble() %>% 
  group_by(cluster) %>% 
  summarize(across(all_of(cols), mean)) %>% 
  arrange(desc(betweenness))

## # A tibble: 4 x 6
##   cluster betweenness in_degree out_degree followers_count friends_count
##   <fct>         <dbl>     <dbl>      <dbl>           <dbl>         <dbl>
## 1 3             4287.      67.5       68.8         310769.         1670.
## 2 1             3568.      75.7       72.9          26351.         1393.
## 3 4             3453.     108.       109.           69698.         1724.
## 4 2             2938.      15.6       17.5          26333.         5449.

Subset

ego_network_subset <- ego_network %>% 
  filter(!is.na(cluster)) %>% 
    mutate(
    out_degree = centrality_degree(mode = "out"),
    in_degree = centrality_degree(mode = "in"),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  ) 

ego_network_subset %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = in_degree), 
                  shape = 21, color = "white", show.legend = FALSE) 

ego_network_subset %>% 
  as_tibble() %>% 
  mutate(label_id = ifelse(
    test = rank(-betweenness) <= 10 |rank(-in_degree) <= 10, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(betweenness, in_degree, color = cluster)) +
  geom_point() +
  ggrepel::geom_label_repel(aes(label = label_id), size = 3)

ego_network_subset %>% 
  group_by(cluster) %>% 
  mutate(label_name = ifelse(
    test = rank(-authority_score) <= 5 | rank(-betweenness) <= 5,
    yes = name,
    no = NA_character_
  )) %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = betweenness), 
                  shape = 21, color = "white", show.legend = FALSE) +
  geom_node_label(aes(label = label_name), 
                  repel = TRUE, alpha = 3/4, size = 3) 

Red de mutuals

Red Personal

edge_list <- list.files(paste0(ego, "_friends_of_friends/"), full.names = TRUE) %>% 
  map(read_rds)

error_index <- edge_list %>% 
  map_lgl(~ any(class(.x) == "try-error")) %>% 
  which()

edge_list <- edge_list[-error_index] %>% bind_rows()

edge_list_mutual <- inner_join(
  edge_list, 
  edge_list %>% rename(from = to, to = from)
  ) %>% 
  filter(from %in% ego_followers$user_id, to %in% ego_followers$user_id) %>%
  filter(from %in% ego_friends$user_id, to %in% ego_friends$user_id) %>%
  filter(from %in% to, to %in% from)

mat <- edge_list_mutual %>% 
  mutate(n = 1) %>% 
  tidytext::cast_sparse(from, to, n) %>% 
  as.matrix()

mat <- mat[colnames(mat), ]

mutual_network <- mat %>% 
  graph_from_adjacency_matrix(mode = "undirected") %>% 
  tidygraph::as_tbl_graph() 

mutual_network

## # A tbl_graph: 1053 nodes and 18582 edges
## #
## # An undirected simple graph with 1 component
## #
## # Node Data: 1,053 x 1 (active)
##   name               
##   <chr>              
## 1 1890827274         
## 2 753376280          
## 3 147705809          
## 4 1113640592842600448
## 5 12542002           
## 6 718862377941397504 
## # … with 1,047 more rows
## #
## # Edge Data: 18,582 x 2
##    from    to
##   <int> <int>
## 1     1     8
## 2     1    14
## 3     1    17
## # … with 18,579 more rows

ego_mutuals_info <- lookup_users(as_tibble(mutual_network)$name, token = sample(token), 1)

ego_mutuals_info <- ego_mutuals_info %>% 
  filter(!protected) %>% 
  select(
    user_id, screen_name, lang, name, location, description,
    ends_with("count"), -starts_with("quote"), 
    -starts_with("retweet"), -reply_count,
    -starts_with("fav")
    ) %>% 
    rename(name = user_id, user_name = name)

mutual_network <- mutual_network %>% 
  inner_join(ego_mutuals_info) %>% 
  rename(name = screen_name, user_id = name) %>% 
  select(name, everything())

## Estadísticas descriptivas

mutual_network <- mutual_network %>% 
  mutate(
    degree = centrality_degree(),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  )

La siguiente gráfica muestra la influencia de cada usuario en Twitter (eje horizontal) vs la influencia de cada usuario dentro de la red personal de amigos (eje vertical)

mutual_network %>% 
  as_tibble() %>% 
  ggplot(aes(followers_count, degree)) + 
  geom_point() 

mutual_network %>% 
  as_tibble() %>% 
  mutate(label_name = ifelse(
    test = rank(-followers_count) <= 15 | rank(-degree) <= 15, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(followers_count, degree)) + 
  geom_point() + 
  ggrepel::geom_label_repel(aes(label = label_name), size = 3)

Clusters

clusters <- igraph::cluster_louvain(graph = mutual_network)

cluster_df <- tibble(cluster = factor(clusters$membership), name = clusters$names) 
  
cluster_df <- cluster_df %>% 
  group_by(cluster) %>% 
  filter(n() >= 10) %>% 
  ungroup()

mutual_network <- mutual_network %>% 
  left_join(cluster_df)

mutual_network %>% 
  as_tibble() %>% 
  arrange(desc(degree)) %>%
  filter(!is.na(cluster)) %>% 
  group_by(cluster) %>%
  filter(rank(-authority_score) <= 30) %>% 
  ggplot(aes(label = name, size = log(degree), color = degree)) + 
  geom_text_wordcloud_area(family = "Avenir Next Condensed") + 
  facet_wrap(~cluster) + 
  labs(title = "Usuarios prominentes en cada cluster") + 
  scale_color_gradient(low = "grey", high = "purple") 

Tamaño de cada cluster:

mutual_network %>% as_tibble() %>% count(cluster)

## # A tibble: 7 x 2
##   cluster     n
##   <fct>   <int>
## 1 1          79
## 2 2          40
## 3 3         170
## 4 4         140
## 5 5         376
## 6 6         235
## 7 7          13

¿Quiénes son los usuarios que funcionan como “puentes”?

mutual_network %>% 
  as_tibble() %>% 
  arrange(desc(betweenness)) 

## # A tibble: 1,053 x 15
##    name  user_id lang  user_name location description followers_count
##    <chr> <chr>   <chr> <chr>     <chr>    <chr>                 <int>
##  1 Erik… 101526… es    C. Erik … "Bogotá… PhD City R…            2585
##  2 psan… 527578… es    Pablo Sa… "Latin … Public Pol…            5891
##  3 spac… 152072… en    Sara Koo… "Kent, … political …            8399
##  4 malb… 482533… es    Mauricio… "Bogotá" Santandere…           39675
##  5 JGar… 585999… es    JavierGa… ""       Profesor C…            2130
##  6 RAKa… 572136… en    Robert A… "Tierra… Colombia p…            9473
##  7 anti… 310297… en    Antipode  ""       Antipode: …           19482
##  8 Ange… 373413… es    Angélica… ""       Ciudadana,…          464194
##  9 regs… 590838… en    Regional… "Bright… Supporting…            8423
## 10 ethe… 296281… es    Ethel Se… "London… MPhil/PhD …             847
## # … with 1,043 more rows, and 8 more variables: friends_count <int>,
## #   listed_count <int>, statuses_count <int>, degree <dbl>, betweenness <dbl>,
## #   authority_score <dbl>, eigen_centrality <dbl>, cluster <fct>

cols <- c("betweenness", "degree", "followers_count", "friends_count")

mutual_network %>% 
  as_tibble() %>% 
  group_by(cluster) %>% 
  summarize(across(all_of(cols), mean)) %>% 
  arrange(desc(betweenness))

## # A tibble: 7 x 5
##   cluster betweenness degree followers_count friends_count
##   <fct>         <dbl>  <dbl>           <dbl>         <dbl>
## 1 7              967.   4.92          15119.         9400.
## 2 3              928.  28.7            3513.         1641.
## 3 5              903.  39.7           14591.         1992.
## 4 6              852.  45.1            2274.         1539.
## 5 4              805.  23.4            1961.          694.
## 6 1              658.  34.9            1435.         1106.
## 7 2              551.  17.0            5194.         2861.

Subset

mutual_network_subset <- mutual_network %>% 
  filter(!is.na(cluster)) %>% 
    mutate(
    degree = centrality_degree(),
    betweenness = centrality_betweenness(directed = TRUE),
    authority_score = centrality_authority(),
    eigen_centrality = centrality_eigen(directed = TRUE)
  ) 

mutual_network_subset %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = degree), 
                  shape = 21, color = "white", show.legend = FALSE) 

mutual_network_subset %>% 
  as_tibble() %>% 
  mutate(label_id = ifelse(
    test = rank(-betweenness) <= 10 |rank(-degree) <= 10, 
    yes = name, 
    no = NA_character_)
    ) %>% 
  ggplot(aes(betweenness, degree, color = cluster)) +
  geom_point() +
  ggrepel::geom_label_repel(aes(label = label_id), size = 3)

mutual_network_subset %>% 
  group_by(cluster) %>% 
  mutate(label_name = ifelse(
    test = rank(-degree) <= 5 | rank(-betweenness) <= 5,
    yes = name,
    no = NA_character_
  )) %>% 
  ggraph("mds") +
  geom_edge_fan(alpha = 1/5, width = 1/5) + 
  geom_node_point(aes(fill = cluster, size = betweenness), 
                  shape = 21, color = "white", show.legend = FALSE) +
  geom_node_label(aes(label = label_name), 
                  repel = TRUE, alpha = 3/4, size = 3) 

Funciones adicionales

readLines("rtweet_functions.R") %>% 
  writeLines()

## 
## # main functions ----------------------------------------------------------
## 
## multi_get_friends <- function(u, token_list) {
##   
##   user_info <- lookup_users(u, token = sample(token_list, 1)[[1]])
##   fc <- user_info$friends_count
##   message("<<", user_info$screen_name, ">> is following ", scales::comma(fc), " users ")
##   
##   if (user_info$protected) stop(call. = FALSE, "The account is protected, we can't get followers.")
##   
##   num_queries <- ceiling(fc / 5000)
##   rl <- rate_limit(token_list, "get_friends")
##   rl <- validate_rate_limit(rl, "get_friends", token_list)
##   
##   index <- get_available_token_index(rl)
##   
##   # Case 0: User doesn't have any friends
##   
##   if (fc == 0) return(tibble(from = character(0), to = character(0))) 
##   
##   # Case 1: Less than 5,000 friends, only call is needed
##   
##   if (fc <= 5e3) {
##     
##     friends <- get_friends(u, token = token_list[[index]])
##     
##   } else {
##     
##     # Case 2: Many calls are needed
##     
##     output <- vector("list", length = num_queries)
##     output[[1]] <- get_friends(u, token = token_list[[index]])
##     
##     for (i in 2:length(output)) {
##       
##       rl <- validate_rate_limit(rl, "get_friends", token_list)
##       index <- get_available_token_index(rl)
##       output[[i]] <- get_friends(u, token = token_list[[index]], page = next_cursor(output[[i - 1]]))
##       
##     }
##     
##     friends <- bind_rows(output) %>% 
##       distinct()
##     
##   }
##   
##   attr(friends, "next_cursor") <- NULL
##   
##   friends %>% 
##     rename(from = user, to = user_id) %>% 
##     mutate(from = user_info$user_id)
##   
## }
## 
## multi_get_timeline <- function(u, n, token_list, home = FALSE) {
##   
##   message(u)
##   rl <- rate_limit(token_list, "get_timeline")
##   rl <- validate_rate_limit(rl, "get_timeline", token_list)
##   
##   index <- get_available_token_index(rl)
##   
##   # Case 0: User doesn't have any posts
##   
##   # what to do?
##   
##   # Should we allow to get all the timeline??? If so, mimic previous function
##     
##   tl <- get_timeline(u, n = n, home = home, token = token_list[[index]])
## 
##   return(tl)
##   
## }
## 
## # multi_lookup_users <- function() {
## #   
## #   
## # }
## 
## 
## # helpers -----------------------------------------------------------------
## 
## validate_rate_limit <- function(rl, q, token_list) {
##   
##   if (is_empty(rl)) {
##     message("Waiting for rate limiting update")
##     Sys.sleep(60)
##     rl <- rate_limit(token_list, query = q)
##     validate_rate_limit(rl, q, token_list) # recursion!
##     
##   }
##   
##   if (all(rl$remaining == 0)) {
##     
##     message("Waiting for token reset in ", round(min(rl$reset), 1), " minutes")
##     Sys.sleep(min(as.numeric(rl$reset_at - Sys.time(), units = "secs")) + 5)
##     rl <- rate_limit(token_list, query = q)
##     validate_rate_limit(rl, q, token_list) # recursion!
##     
##   }
##   
##   rl
##   
## }
## 
## get_available_token_index <- function(rl) {
##   
##   env <- rlang::caller_env()
##   available_token <- rl$remaining > 0
##   index <- which(available_token)[[1]]
##   env$rl[index, ]$remaining <- rl[index, ]$remaining - 1  # this modifies the rl obj in the parent frame
##   return(index)
##   
## }

theme_custom

## function (base_family = "Avenir Next Condensed", fill = "white", ...) {
##     theme_minimal(base_family = base_family, ...) %+replace% 
##         theme(plot.title = element_text(face = "bold", margin = margin(0, 
##             0, 5, 0), hjust = 0, size = 13), plot.subtitle = element_text(face = "italic", 
##             margin = margin(0, 0, 5, 0), hjust = 0), plot.background = element_rect(fill = fill, 
##             size = 0), complete = TRUE, axis.title.x = element_text(margin = margin(15, 
##             0, 0, 0)), axis.title.y = element_text(angle = 90, 
##             margin = margin(0, 20, 0, 0)), strip.text = element_text(face = "italic", 
##             colour = "white"), strip.background = element_rect(fill = "#4C4C4C"))
## }